Fuzzy Inference System Optimized by Genetic Algorithm for Robust Face and Pose Detection

Abstract

A human face detection method for color images is presented in this paper, which is pose, size and position independent, and has the priority of classifying detected faces in three groups: frontal, near frontal and profile, according to their pose. This system is a fuzzy rule base one, optimized by genetic algorithm. In the first stage, skin color regions are selected in the input image. Within each skin area, lip pixels and ear texture are searched, and applied as features to identify face candidates in the skin regions. Summarizing all obtained information along by skin region shape and lip area position relative to skin area, four inputs are computed for fuzzy inference system, and face areas as well as their poses would be introduced. The proposed method is tried on various databases, including HHI, Champion, Caltech, Bao and IMM databases. Achieved results show a remarkable detection rates compared to other methods, for various face poses. 96.8%, 95.3% and 87.8% correct detection rates are achieved, respectively for frontal, near frontal and profile face images over 1298 face image samples.